Modeling bat activity across the fire- managed landscape of Mammoth Cave Nat’l Park using remotely-sensed forest canopy data Dodd, L.E. 1, M.J. Lacki 1, N.S. Skowronski 2, M.B. Dickinson 2, & L.K. Rieske 3 1 Forestry Department, University of Kentucky 2 Northern Research Station, US Forest Service 3 Entomology Department, University of Kentucky

Remote Sensing & Wildlife Large-scale patterns 1 Large-scale patterns 1 Feasibility Feasibility Necessity Necessity 1 Vierling et al Frontiers in Ecology & the Environment 6: Hudak et al Remote Sensing 1:

Bats at Mammoth Cave Variable foraging & habitat use across species 1 Variable foraging & habitat use across species 1 Prey availability & forest canopy structure Prey availability & forest canopy structure White-nose syndrome White-nose syndrome Now at Mammoth Cave; changing predator-prey dynamics? Now at Mammoth Cave; changing predator-prey dynamics? 1 Swartz et al Pp in: Bat Ecology. Lacki et al Pp. 83–128 in: Bats in Forests: Conservation and Management

Core Hibernacula Burn Areas Mammoth Cave Nat’l Park

Core Hibernacula Burn Areas Survey Transects, Aug 2010 onward Mammoth Cave Nat’l Park

Core Hibernacula Burn Areas Survey Transects, Aug 2010 onward Mammoth Cave Nat’l Park

Methods

Methods

Methods Bat Activity Zero-crossing acoustic surveys Zero-crossing acoustic surveys Spanning (still ongoing) Spanning (still ongoing) 170 nights (1,086 detector / nights) 170 nights (1,086 detector / nights) Emphasis on April-May, Aug-Oct Emphasis on April-May, Aug-Oct

Methods Bat Variables Echoclass v Echoclass v High freq (> 34 kHz) High freq (> 34 kHz) Low freq (≤ 34 kHz) Low freq (≤ 34 kHz) Feeding buzzes Feeding buzzes 1 USFWS. Indiana Bat Survey Guidance.

Methods Bat Variables Echoclass v Echoclass v High freq (> 34 kHz) High freq (> 34 kHz) Low freq (≤ 34 kHz) Low freq (≤ 34 kHz) Feeding buzzes Feeding buzzes Total Activity 1 USFWS. Indiana Bat Survey Guidance.

Methods LiDAR Survey Figure by Renslow LiDAR = Light Detection and Ranging LiDAR = Light Detection and Ranging Discrete-return scanning LiDAR 1 Discrete-return scanning LiDAR ,600 nm wavelength 900-1,600 nm wavelength > 4 pulses / m² > 4 pulses / m² 1 Skowronski et al Remote Sensing of Environment 108:

Methods LiDAR Survey Figure by Renslow LiDAR = Light Detection and Ranging LiDAR = Light Detection and Ranging Data collected Fall 2010 (leaf-off) via fixed-wing aircraft Data collected Fall 2010 (leaf-off) via fixed-wing aircraft

Methods LiDAR Variables What scale is meaningful? What scale is meaningful?

Methods LiDAR Variables 1 Lesak et al Remote Sensing of Environment 115: Laser returns across over-, mid-, & understory strata 1 Laser returns across over-, mid-, & understory strata 1 Under Mid Over

Methods LiDAR Variables 15 m Laser returns across over-, mid-, & understory strata 1 Laser returns across over-, mid-, & understory strata 1 15 m radii around survey points 1 15 m radii around survey points 1 1 Lesak et al Remote Sensing of Environment 115: Under Mid Over

Methods LiDAR Variables Strata (absolute & relative) Strata (absolute & relative) Over-, mid-, & understory Over-, mid-, & understory Determining canopy shape Determining canopy shape Mid:Over, Under:Mid, & Under:Over Mid:Over, Under:Mid, & Under:Over

Methods LiDAR Variables Strata (absolute & relative) Strata (absolute & relative) Over-, mid-, & understory Over-, mid-, & understory Determining canopy shape Determining canopy shape Mid:Over, Under:Mid, & Under:Over Mid:Over, Under:Mid, & Under:Over Gap Index Gap Index Percentage of pixels with no laser returns >3 m height Percentage of pixels with no laser returns >3 m height

ANOVAs for site, temporal, & fire effects ANOVAs for site, temporal, & fire effects Analysis & Results

ANOVAs for site, temporal, & fire effects ANOVAs for site, temporal, & fire effects Multiple linear regressions relating activity to forest veg Multiple linear regressions relating activity to forest veg Response variables: Response variables: - high freq pulses - low freq pulses - feeding buzzes - high freq pulses - low freq pulses - feeding buzzes Predictive models: Predictive models: - understory - midstory - overstory - “total” clutter - understory - midstory - overstory - “total” clutter Model selection using Akaike’s Information Criterion Model selection using Akaike’s Information Criterion Models in SAS 9.0, then protocol of Burnham & Anderson 1 Models in SAS 9.0, then protocol of Burnham & Anderson 1 1 Model Selection & Multimodal Inference, 2 nd Edition Analysis & Results

Site, Season, & Annual Effects

Unsurprisingly, lots of variation across sites! Unsurprisingly, lots of variation across sites!

Site, Season, & Annual Effects Far fewer replicates (n = 6)

Site, Season, & Annual Effects

High Freq: F 2,1083 = 3.6, P = 0.02 Low Freq: F 2,1083 = 29.2, P < 0.01 AAB ab c

Site, Season, & Annual Effects

A B B a bab No apparent WNS impacts in 2012… No apparent WNS impacts in 2012… High Freq: F 2,1083 = 4.1, P = 0.02 Low Freq: F 2,1083 = 3.0, P = 0.05

Site, Season, & Annual Effects

Sites surveyed in Fall 2010 Sites surveyed in Fall 2010

Effect of Fire Core Hibernacula Burn Areas Survey Transects, Aug 2010 onward Repeated visits have Repeated visits have created a time-series… created a time-series…

As with sites, lots of variation across years since burn! As with sites, lots of variation across years since burn! Effect of Fire

Far fewer replicates (n ~ 5)

Effect of Fire Far more replicates (n > 200 each)

Effect of Fire

High Freq: F 2,1083 = 11.1, P < 0.01 Low Freq: F 2,1083 = 1.0, P = 0.37 A BB Effect of Fire

A BB Feeding Buzzes F 2,1083 = 11.1, P < 0.01 Effect of Fire

Core Hibernacula Burn Areas Survey Transects, Aug 2010 onward

Effect of Fire Analyses pending

Effect of Fire Analyses pending

Effect of Fire Analyses pending

Effect of Fire Relation to Canopy Clutter Data within first year of burning ( ) Data within first year of burning ( )

High frequency echolocators negatively High frequency echolocators negatively associated a cluttered understory associated a cluttered understory

Low frequency echolocators negatively Low frequency echolocators negatively associated with a cluttered understory & overstory associated with a cluttered understory & overstory

Feeding buzzes negatively associated with a cluttered Feeding buzzes negatively associated with a cluttered understory & overstory; positively associated with canopy gaps understory & overstory; positively associated with canopy gaps

Low Frequency Bat Activity Predictive Map – Total Clutter

Discussion & Implications Activity lower at burned sites, but models suggest activity is also Activity lower at burned sites, but models suggest activity is also positively related to less-cluttered canopy conditions… positively related to less-cluttered canopy conditions… High frequency & low frequency echolocators both respond to High frequency & low frequency echolocators both respond to differences in clutter… And are relatable to management! differences in clutter… And are relatable to management! How does this relate to longer-term fire management plans? How does this relate to longer-term fire management plans?

Funding Funding – Joint Fire Science Program NPS Personnel NPS Personnel – Dr. Rick Toomey – Steve Thomas – Shannon Trimboli Tech Support! Tech Support! – Tracy Culbertson – Stella James – Klint Rose – Jennifer Winters Thanks!